What Are Spot Instances? Affordable Compute in the Cloud

Spot instances are unused cloud compute capacity offered at steep discounts, typically 60-90% less than on-demand prices. According to AWS, Spot Instances save an average of 70-80% compared to on-demand EC2 pricing. The trade-off: the cloud provider can reclaim these instances with minimal notice when demand for that capacity rises. This makes spot instances ideal for fault-tolerant, interruptible workloads.

Key Takeaways

• Spot instances save 60-90% compared to on-demand pricing (AWS).
• Instances can be interrupted with as little as 2 minutes notice on AWS.
• Best for batch jobs, CI/CD, data processing, and stateless applications.

[INTERNAL-LINK: cloud cost optimization services → pillar page]

How Do Spot Instances Work?

Cloud providers maintain large pools of compute capacity. When customer demand doesn't fully consume that capacity, providers offer the surplus as spot instances at reduced prices. On AWS, you request Spot Instances by specifying instance types, availability zones, and a maximum price you're willing to pay. If capacity is available at or below your price, you get the instances.

When the provider needs the capacity back, your instances receive a termination notice. AWS gives a 2-minute warning. Azure provides a 30-second eviction notice for Spot VMs. GCP Preemptible VMs get a 30-second warning, while Spot VMs on GCP receive varying notice periods. Your application must handle these interruptions gracefully.

What Are Spot Instances Called on Each Provider?

Each cloud provider brands spot capacity differently. Azure calls them Spot Virtual Machines, with discounts of up to 90% for Linux and 80% for Windows workloads. AWS uses the name EC2 Spot Instances. GCP offers both Spot VMs (newer, recommended) and Preemptible VMs (legacy, limited to 24-hour runtime). Despite the naming differences, the concept is identical across providers.

Pricing mechanisms vary slightly. AWS uses a dynamic market price that fluctuates based on supply and demand. Azure lets you set a maximum price or use the pay-as-you-go rate. GCP charges a dynamic price that's typically 60-91% below on-demand. All three providers charge per second of actual usage.

Which Workloads Are Good Candidates for Spot Instances?

Spot instances work best for workloads that can tolerate interruptions. Datadog's infrastructure research shows that organizations using spot instances allocate them primarily to batch processing (38%), CI/CD pipelines (27%), and data analytics (22%). These workloads share a common trait: they can checkpoint progress and resume after an interruption without data loss.

Good candidates: batch data processing, machine learning training, CI/CD build jobs, rendering workloads, stateless web servers behind load balancers, and big data analytics. Poor candidates: databases, stateful applications without checkpointing, single-instance workloads with no failover, and latency-sensitive real-time services.

How Do You Reduce the Risk of Spot Interruptions?

Diversification is the primary strategy. Request capacity across multiple instance types, sizes, and availability zones. AWS Spot Fleet and EC2 Auto Scaling groups support this automatically. If one instance type in one zone gets reclaimed, others continue running. A diversified spot strategy reduces interruption rates to below 5% for most workloads.

Combine spot with on-demand or reserved instances using a mixed strategy. Run your baseline capacity on reserved instances and burst capacity on spot. AWS Auto Scaling groups support mixed instance policies that automatically balance between on-demand and spot. This hybrid approach captures spot savings without risking complete workload failure.

[INTERNAL-LINK: cost visibility for spot tracking → /blogs/cloud-cost-visibility-transparency-guide/]

Frequently Asked Questions

Can I use spot instances for production workloads?

Yes, with careful architecture. Stateless applications behind load balancers can run on spot instances if the auto-scaling group maintains minimum capacity across on-demand and spot. When a spot instance is reclaimed, the load balancer routes traffic to remaining instances while a replacement launches. Many organizations run 50-70% of their web tier on spot instances.

How much do spot instances actually save?

Savings depend on instance type and region. Common instance types like m5.large on AWS typically offer 65-75% discounts. Less popular instance types or newer generations may offer 80-90% discounts. Track your actual spot savings in AWS Cost Explorer or equivalent tools to measure real impact against cloud cost optimization targets.

What happens to my data when a spot instance is interrupted?

Instance storage (ephemeral) is lost when a spot instance terminates. EBS volumes persist if configured to survive termination. Always store important data on persistent storage (EBS, S3, or equivalent). Use the interruption notice to flush in-progress work to persistent storage before the instance shuts down.